Key Issues Of Question Understanding In Community Question Answering System

The rapid development of Community Question Answer(CQA)System attracts the attention of many scholars coming from both research and industry areas.Currently,it has also become an important branch of Natural Language Processing(NLP)and Information Retrieval(IR).The questions from the users are usually very short and convey limited information.Furthermore,the expression of the same question can be varied.So the existing retrieval tools can not understand the nature language very well,and their search results are not satisfactory.During the past period,the community question answer systems have accumulated a large number of question-answer pairs,such as ”Yahoo! Answer”,”Sina i Ask”and so on.How to enhance the ability of question understanding in these systems using the historical data to provide better service has become one of the research hotspots.Using Yahoo! Answer's Q&A corpus,this paper constructs a question understanding platform for Community Question Answer System.After classifying the proposed question into different categories,the platform further identifies the intention of the question.Then the platform finds the top-N best match historical questions considering category,semantics and the intention from the existing knowledge base,i.e.,the accumulated historical data set.Finally,incorporating the known answers corresponding to the match questions,the platform further determines the closest historical question and recommends the corresponding answer to the user.In the process of constructing the platform,three key issues are studied in this paper.Specific research contents include:(1)In order to better understand the semantics of the proposed question and reduce search space,question classification task is studied.We propose a novel approach combining lexical,syntactic and dependency information to this task.The experiment results on the UIUC corpus show that our proposed approach can classify the question efficiently.(2)In order to better understand the intention of the question,we propose a keyword-based question focus recognition approach.The effectiveness of the approach is also verified on the Yahoo! Answer corpus.(3)In order to find the closest historical question in the knowledge base,we propose a translation model-based question mapping approach.The experiment results on the Yahoo! Answer corpus show the effectiveness of our approach.